While Vision-Language-Action (VLA) models have revolutionized autonomous driving by unifying perception and planning, their reliance on explicit textual Chain-of-Thought (CoT) leads to semantic-perceptual decoupling and perceptual-symbolic conflicts. Recent shifts toward latent reasoning attempt to bypass these bottlenecks by thinking in continuous hidden space. However, without explicit intermediate constraints, standard latent CoT often operates as a physics-agnostic representation. To address this, we propose the Latent Spatio-Temporal VLA (LaST-VLA), a framework shifting the reasoning paradigm from discrete symbolic processing into a physically grounded Latent Spatio-Temporal CoT. By implementing a dual-feature alignment mechanism, we distill geometric constraints from 3D foundation models and dynamic foresight from world models directly into the latent space. Coupled with a progressive SFT training strategy that transitions from feature alignment to trajectory generation, and refined via Reinforcement Learning with Group Relative Policy Optimization (GRPO) to ensure safety and rule compliance. \method~setting a new record on NAVSIM v1 (91.3 PDMS) and NAVSIM v2 (87.1 EPDMS), while excelling in spatial-temporal reasoning on SURDS and NuDynamics benchmarks.
@article{arxiv.2603.01928,
title = {LaST-VLA: Thinking in Latent Spatio-Temporal Space for Vision-Language-Action in Autonomous Driving},
author = {Yuechen Luo and Fang Li and Shaoqing Xu and Yang Ji and Zehan Zhang and Bing Wang and Yuannan Shen and Jianwei Cui and Long Chen and Guang Chen and Hangjun Ye and Zhi-Xin Yang and Fuxi Wen},
journal= {arXiv preprint arXiv:2603.01928},
year = {2026}
}